[ImageJ-devel] Threshold command in IJ2

[Lee Kamentsky]

We ran into the same requirement here. We had brightfield images where the
background was uniform and the foreground was lighter or darker. A simple
median worked well to stabily identify the background, but I was not able
to come up with a good way to identify the high and low cutoffs. My first
guess was to separate the pixel values into those that were above the
median and those that were below and then use the Otsu algorithm on each
group of pixel values to find the low and high cutoffs; that overestimated
the background for my images. In the end, we just used a fixed increment
and decrement from the median which worked well enough for our images. My
instinct is that a similar technique should work and should be more
reliable than something based on percentiles - the assumption made by
percentiles is that there are a certain number of background pixels and
that will vary from one image to the next.

I think a successful method might make a good paper, haven't done a
literature search of the problem though, maybe something is out there.

--Lee


On Thu, Jul 25, 2013 at 12:51 PM, Aivar Grislis <grislis at wisc.edu> wrote:

> In the Threshold command we should support threshold methods that
> automatically exclude both low and high values.
>
> All of the current threshold methods separate figure from background by
> coming up with a single threshold value. The figure is then determined to
> be pixels with values from the minimum value to the threshold value and
> background threshold value to maximum value.  If the 'dark background' box
> is checked these ranges are swapped.  In the user interface and internal
> representation rather than having a single threshold value there is a
> minimum and maximum range that the user can tweak.
>
> As an example, for images that consist of data for each pixel that is
> expected to have a near-normal distribution, such as fitted lifetime
> images, it's useful to have statistics-based threshold methods that exclude
> outliers based on mean and standard deviation or else median and quartile
> ranges.  (I like to use Tukey's rule: http://www.edgarstat.com/**
> tukeys_outliers_help.cfm<http://www.edgarstat.com/tukeys_outliers_help.cfm>)  Are there any other useful algorithms we could implement to exclude low
> and high values in ordinary images?
>
> Thanks,
> Aivar
>
>
> ______________________________**_________________
> ImageJ-devel mailing list
> ImageJ-devel at imagej.net
> http://imagej.net/mailman/**listinfo/imagej-devel<http://imagej.net/mailman/listinfo/imagej-devel>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://imagej.net/pipermail/imagej-devel/attachments/20130725/7160903d/attachment.html>